A beginner’s guide to stereo-derived DEM production and analysis using
ISIS, ASP, and ArcMap
v. 1.1
Teemu Öhman
LPI, June 2013
T. Öhman
6/21/2013
Table of contents
Preface, or things to note before you start........................................................................................3
1. Finding the right images...............................................................................................................4
1.1.
CTX ......................................................................................................................................4
1.2. HRSC ...................................................................................................................................4
1.3. LRO NAC ............................................................................................................................5
2. Processing the images with ISIS ..................................................................................................5
2.1.
Cubing the raw .....................................................................................................................5
2.2.
Adding some spice ...............................................................................................................6
2.3.
Radiometric calibration ........................................................................................................6
2.4.
Creating a map template.......................................................................................................6
2.5.
Projecting the cubes .............................................................................................................7
2.6.
Giving the final touches to LRO NACs for ArcGIS/ArcMap..............................................7
3. Creating DEMs using ASP...........................................................................................................8
3.1.
Crop before you start............................................................................................................8
3.2.
Running stereo......................................................................................................................9
3.3.
Changing the correlation parameters..................................................................................10
3.4.
Creating the DEMs.............................................................................................................12
3.4.1.
The Moon .......................................................................................................................12
3.4.2.
Exporting the DEM to ArcGIS.......................................................................................12
3.4.3.
Visualising the DEM in ASP .........................................................................................12
3.4.4.
Martian DEMs and the areoid ........................................................................................14
4. Processing and analysing the DEMs in ArcMap........................................................................16
4.1.
Importing and re-projecting the data..................................................................................16
4.2.
Subtracting the areoid.........................................................................................................17
4.3.
Displaying the DEM...........................................................................................................17
4.4.
Georeferencing ...................................................................................................................18
4.5.
Analysing the DEM and exporting the data .......................................................................19
Acknowledgements ........................................................................................................................19
Appendix 1. Some additional links and references ........................................................................20
Appendix 2. Some very basic UNIX/LINUX commands..............................................................22
Appendix 3. An example of a stereo parameter file for processing LRO NAC stereo pairs .........23
Appendix 4. An example of a stereo parameter file for processing CTX stereo pairs...................26
Appendix 5. An example of a stereo parameter file for processing HRSC stereo pairs ................29
DEMs for dummies
Page 2 of 30
T. Öhman
6/21/2013
Preface, or things to note before you start
The following is a brief beginner’s guide to producing and analysing planetary (the Moon and Mars,
but the basics are the same for the other heavenly bodies as well) stereo image-based digital
elevation models (DEMs) with USGS Integrated Software for Imagers and Spectrometers (ISIS),
NASA Ames Stereo Pipeline (ASP), and ESRI ArcGIS (Geographic Information System). The
procedures have been tested with ISIS 3.3.0.3646 and ASP 1.0.5 (both running on lpi24), and
ArcGIS 10.0. More recent versions of all three exist, but this set-up has the benefit that I have
proven it works. I am mostly describing Martian DEMs, but will mention lunar-specific commands
whenever necessary. The example files mentioned here are files that I have successfully used in
creating the DEMs, so this is thoroughly tested in practice.
An important thing to note is that these DEMs described here do not use ground control points and
bundle adjustment, and therefore do not provide accurate absolute elevations. Hence, these
elevations should not be compared between different DEMs. Within a single DEM, however, the
elevations have on all tests that I have done been consistent. Ground control points can at least
theoretically be used with substantially more ISISing and ASPing, and it is quite time-consuming
and difficult. Even the ASP developers admit that. With the current set-up, I have never managed to
do it. Nevertheless, peer-reviewed research has been published using this simplified approach
(Lefort et al. 2012).
Another very significant thing to realise is that ASP does not allow you to create DEMs from
mosaics, only from single stereo pairs. For mosaic DEMs, you have to use Socet Set or other similar
commercial software. In the future, it would also be useful to look into ASP 2.x, which is already
available, and is claimed to be substantially faster.
The approach here is from a dummy, to a dummy. Even if you are not going to be using DEMs, the
first and second parts of this short set of instructions can be used as a guide for extreme beginners to
find the correct set of images and to process them with ISIS. Some of the ArcMap analysis part
might be handy as well.
Although there are bound to be inconsistencies, I have made some attempts for a consistent style to
make this easier to follow by just glancing. Commands that are supposed to be written exactly like
that in the command line are indented and written in arial. In the case of ASP, these are usually
followed by things like output prefixes, which are written in times new roman italics. Options in the
ISIS GUI that are mentioned in the text are also given in times new roman italics, and it is used for
emphasis in a few other places as well. In case you don’t want to use the GUI, it probably helps that
the actual ISIS options you would use in the command line are written with CAPITALS in times new
roman italics. When you need to type something in ISIS GUI, it is given in “quotation marks in
italics”. ASP stereo parameter file text, mostly given in the appendices, is written in courier
new.
DEMs for dummies
Page 3 of 30
T. Öhman
6/21/2013
1. Finding the right images
There is a confusing number of places to download planetary images and other data. Some of the
places provide the same data as other places, some don’t, even if they are supposed to. For the
purposes of creating DEMs, I have found the following places to be the best for me.
1.1. CTX
The best Martian DEMs that I have encountered are done using MRO CTX (Mars Reconnaissance
Orbiter Context Camera) images, with a resolution of about 6 m/px. The stereo coverage is not
extensive, and some apparent stereo pairs have in practise too large or small differences in
illumination or viewing geometries to work well. To get an idea if a stereo pair has the potential of
producing a good DEM, it helps and makes things faster if you are able to see the browse versions
of the stereo pairs on screen in 3D without stereo glasses. If the 3D effect looks good, the chances
are that the DEM will too. Some people cannot do it, but most people can learn it – either crossvision or parallel-vision – with little practice.
For CTX, the best place to obtain the images is the MRO CTX image map:
http://global-data.mars.asu.edu/bin/ctx.pl
Select the image, and download the PDS source EDR raw image. The correct image format should
have
a
name
like
G20_025970_2217_XN_41N102W.IMG
or
G20_025904_2209_XN_40N102W.IMG from the eastern flank of Alba Mons.
1.2. HRSC
MEX HRSC (Mars Express High Resolution Stereo Camera) stereo channel images are obviously
very good for DEMs as well, although they usually require more work to produce nice DEMs with
ASP. Finding them is also more difficult. The place to go is the HRSCview data explorer:
http://hrscview.fu-berlin.de/cgi-bin/ion-p?page=entry2.ion
Select the HRSC Archive DTM resolution map. Once there, select the image you want using the
very confusing map interface. When there are two or more images available, the way to choose the
image further down the list for downloading is to select it on the drop-down menu and then hit the
Update view button. This updates the almost hidden link to the actual download page at the bottom
of the page.
At the download page, the images to download are in the Download (level 2) PSA (PDS format)
column. The stereo channel images have names like H0068_0009_S12.IMG and *S22.IMG, also
from the eastern flank of Alba Mons.
If DLR (German aerospace center, or Deutsches Zentrum für Luft- und Raumfahrt) has produced a
DEM of this stereo pair, it is available from this page as well. The areoid-referenced DEMs are the
ones that give elevation data that is referenced to MOLA, so you should directly get the
approximately same elevation numbers from the DLR areoid-referenced DEM, MOLA points, and
MOLA DEM without any additional fiddling. In this case, the level-4 areoid-referenced DLR DEM
made from the images mentioned above is H0068_0009_DA4.IMG.
DEMs for dummies
Page 4 of 30
T. Öhman
6/21/2013
1.3. LRO NAC
For the Moon and Lunar Reconnaissance Orbiter Narrow Angle Camera images, or LRO NACs
(and Wide Angle Camera Images, or WACs), I would start with the LROC QuickMap:
http://target.lroc.asu.edu/q3/
You can select the NAC from there, or if you know the coordinates for your images, you can go
directly to the LROC Image Search page:
http://wms.lroc.asu.edu/lroc/search
And from there, after setting your search parameters, you will get to the download page. What you
want to download is the EDR, not the CDR! The correct image file should have a name like
M152451994LE.IMG or M152445210LE.IMG, as is the case here for crater Eimmart A. One thing
that might be good to know is that those numbers in LROC names are seconds (mission elapsed
time, I suppose). You most likely would want to get both the left (*LE.IMG) and the right
(*RE.IMG) NACs of both of your stereo pairs, but that does of course depend on the purpose and
the stereo coverage. And no, the tiny little zone of overlap between the two NACs (left and right)
does not work as a stereo pair. You’ll just use that overlap to mosaic them accurately in Photoshop.
2. Processing the images with ISIS
First of all you need to have a user account on lpi24, and you need to be able to establish an SSH
Secure Shell connection to lpi24. These have to be done by the IT-guys. Once that is done and you
have logged in, hit the New File Transfer Window button, then create New Folder, and copy the
files to this folder (just drag and drop). Then you can start ISISing.
ISIS can be run either from the graphic user interface (GUI) or directly from the command line. To
start with, GUI is easier, and that is what’s being used here. There is no “ISIS GUI”, but each
program in ISIS has its own GUI, which you start by typing the name of the program on the
command line. After becoming more familiar with ISIS, some (if not all) of the following stages are
a bit faster to do directly from the command line. One thing that command line usage is good for is
batch processing, which is something to look into particularly if you are about to process large
amounts of images with the same set of parameters.
2.1. Cubing the raw
The first step is to convert the raw images to cubes. These are done with mission-specific
commands. For CTX, HRSC, and LRO NAC the commands are
mroctx2isis
hrsc2isis
and
lronac2isis
DEMs for dummies
Page 5 of 30
T. Öhman
6/21/2013
respectively. In the GUI, simply select your raw (*.IMG) image in the FROM-field, and potentially
rename it to something more understandable in the TO-field. Adding .cub to the file name is not
necessary, as ISIS does it automatically. Hit the green button to run it.
For other useful datasets, for example Kaguya TC, you would use raw2isis, but that is somewhat
more complicated and not dealt with here (although I emphasize that Kaguya TC is a fantastic but
overlooked dataset with a resolution of about 7.4 m/px and, therefore, it is extremely useful between
the 60–100 m/px LRO WACs and 0.4–1.2 m/px LRO NACs).
2.2. Adding some spice
Adding SPICE data (spacecraft location and attitude, camera pointing data, etc. See the link in
Appendix 1.) to the images can be done either now or after radiometric calibration. For CTX and
LRO NAC (and LRO WAC as well) you simply run
spiceinit
and keep all the settings at defaults. However, for HRSC you have to select the CKPREDICTED
(Use predicted pointing kernels if available) option. Note that for very recent images, the spice
kernels on lpi24 may not always be up-to-date. In that case, you need the IT-guys to update them.
2.3. Radiometric calibration
Radiometric calibration does very little visible things to the images, but it should be run anyway.
It’s perhaps good to mention here that for CTX and LRO NAC images, photometric calibration is
totally pointless. For making sensible-looking LRO WAC mosaics it is vital, but that is another
story and not dealt with here. For LRO NAC and CTX the radiometric calibration is simply run
with commands
lronaccal
and
ctxcal
and keeping all the settings at defaults. For HRSC, ISIS does not have a radiometric calibration
routine, and things work just fine without it.
2.4. Creating a map template
This step is only required when you start a new project, because you will mostly be using the same
template in all of your images (unless you are studying, for example, both poles and equatorial
regions). Start creating the template by typing:
maptemplate
Give your map a name (again, sensible and descriptive names are advisable, because you are likely
going to be using several map templates during different stages of different projects and without
good names you will end up using a wrong template at some point). ISIS will add .map to the
template name. Skip the FROMLIST-field. Choose the projection and projection parameters
DEMs for dummies
Page 6 of 30
T. Öhman
6/21/2013
according to your liking. Central longitude (CLON) is always required, Central latitude (CLAT) in
many cases.
In TARGOPT under Target parameters, choose User input (USER). For TARGETNAME, type either
“moon” or “mars”. In the most straightforward case you can keep all the remaining settings at
defaults. Alternatively, you can define the latitude and longitude extent and the resolution of your
map, but just for doing a simple DEM it is not necessary. If you plan to do an image mosaic (not
DEMs), then it is a good idea to set the resolution already here in the map template so you don’t
have to worry about it while projecting the individual images.
2.5. Projecting the cubes
Doing the map projection is the time-consuming part of the basic ISISing. Start it by typing:
cam2map
In the FROM-field, select your calibrated cube, and in the MAP-field choose the map template you
just made. In the TO-field I would give the file a new, again sensible and descriptive name so you
know what you have done to the image just by looking at its name.
Under Output Map Resolution you have several useful options, depending on the case. You can just
use the default setting Compute resolution from input cube (CAMERA), but that’s usually not the
best way. If your final goal is an image mosaic with a consistent resolution, then the easiest thing to
do is to define the resolution in the map template, and then choose Read resolution from input map
file (MAP). If you just want to keep the best possible resolution but make a mosaic of LRO NAC
left and right images, then choose Get resolution from user in meters per pixel (MPP). As an aside,
if you only need an LRO NAC left and right mosaic without any geographic information, you’ll get
better results and you’ll get them much faster if you just do it in Photoshop rather than trying any
mosaicking options of ISIS.
The next options (Output Map Ground Range and Longitude Seam Options) are usually best kept at
defaults, but the choice for the last one, Options, depends on the images you use. Nearest neighbor
is the fastest, and the one that I have almost always used for LRO NACs and CTXs. However, for
low-resolution things like the areoid (see below), a bi-linear interpolation can often give the best
result.
If you haven’t already looked at the intermediate versions of your cube, you probably want to look
at this final product:
qview
In qview it’s useful to verify that no mistakes were made during the map projection. Clicking the
icon with footprints opens the Advanced Tracking window. Move the mouse over the image and
check that the coordinates change the way they ought to.
2.6. Giving the final touches to LRO NACs for ArcGIS/ArcMap
There are a number of ways to export cubes so they can be used in ArcGIS. You can try isis2std or
various gdal scripts to create geotiffs or jpeg2000s, but the best way, however, is not to export them
in different file formats at all, because ArcGIS 10 can handle most cubes without problems. LRO
DEMs for dummies
Page 7 of 30
T. Öhman
6/21/2013
NACs are problematic though, because with the processing done so far, NACs in Arc end up having
large black boundaries. For example a jp2-file exported with gdal will be a rectangle with
horizontal and vertical boundaries in ArcMap, so if the original image is taken/projected so that it
runs “obliquely” in your ArcMap project, you will end up with much of your image being black. To
avoid that, you need to delete some of the groups in the image header. Run
editlab
and choose Delete group (DELG). Under the Input Values, type “kernels” in GRPNAME. Do the
same for “radiometry”. After this, your cubes are ready to be used in ArcMap. Other cubes
(including LRO WACs, although they require additional processing steps not described here) work
just fine without this step.
If you are not going to be using DEMs, you can jump straight into section 4 where basic
ArcMapping is discussed.
3. Creating DEMs using ASP
3.1. Crop before you start
Finishing all of the above steps (except for deleting the groups of LRO NACs described in section
2.6) are a pre-requisite for creating a good DEM with ASP. So what you need is a stereo pair with
both images having the same resolution and same projection. What you also need is to know which
image is the “left eye” image and which is the “right eye” image. This is easy if you can see stereo
without using stereo glasses, but if not, then you just have to try to figure it out by looking at the
different perspectives of the images. If everything else fails, just try to create a DEM, and if you
don’t get a sensible result despite everything being alright, you most likely have left and right
reversed.
Creating DEMs is very time-consuming (ASP 2 is, at least they say so, significantly faster), so for
the initial tests you need to have your images heavily cropped. Don’t use cubes larger than 10 Mb
for testing, and smaller ones will do. It is worthwhile to make the cropped image as representative
as possible, so if your image has different terrains (e.g., flat, fairly featureless areas, as well as
craters) try to include them both, because your initial set of parameters might work fine for the
rougher topography, but utterly fail in the smoother parts of the image.
Cropping in ISIS is simply done by:
crop
FROM and TO-fields are by now obvious, so they’re just the file names. Sample and Line
Extraction are a bit more complicated. Samples increase from left to right (so can be thought of as
columns), and lines from top to bottom (so the spreadsheet analogue is rows). In ISIS, the top left
corner of the top left pixel has coordinates 0.5 0.5! That is actually an important piece of
information.
DEMs for dummies
Page 8 of 30
T. Öhman
6/21/2013
When doing the cropping of a stereo pair in practice, I have found it best to have both of the images
open in qview at the same time, locating a good place for cropping, and then using the Advanced
Tracking to get the upper left coordinates (sample and line) of the to-be-the-cropped-image,
obviously for both images of the stereo pair. These upper left corner coordinates are what you are
going to type into the SAMPLE and LINE fields in the crop GUI.
While still in qview, you should also record the sample and line of the lower right corner of your tobe-the-cropped-images. In crop, the NSAMPLES and NLINES fields are the number of samples and
lines of the cropped image. So to get the numbers of lines and samples, subtract the upper left
values of your to-be-the-cropped-image from the lower right ones. The better your left and right
images match, the faster your processing times are, and the better your DEM (if the match is poor,
you need a larger search window to find the matching pixels, which makes processing slow, and
you may end up in a situation where ASP does not find a match, resulting in a hole or a mismatch).
It is normally easy to find matching pixels in the cropping stage within a couple of pixels. Note also
that you have to do this cropping at least twice for each image: first to create the small image for
testing, and after you have found the correct parameters, you need to crop the full (or as large as you
need) images to be as good matches as possible.
SINC and LINC in crop are the sample and line increment. The default values are 1, so the
resolution of the cropped image will be the same as the original resolution. If you set SINC to 2,
only every second sample will be used, and so forth. In the case of high-resolution data (like CTX
and LRO NAC), you will often need to fiddle with these, because full resolution DEM files become
too large to handle, or ASP just doesn’t produce good DEMs with full resolution. The best sampling
rate varies in each case.
3.2. Running stereo
Unlike ISIS, ASP has no GUI, but luckily the number of commands needed for the basic use is very
limited. First of all, you obviously need to have the stereo parameter file in the same folder where
you have your cubes and where you are running ASP. If you don’t specify which stereo parameter
file you will use, the program will look for a file called stereo.default. You can always keep this
name for your stereo parameter file, or you may have a large number of slightly modified versions,
so if the new version results in a worse DEM than in the previous run, you can easily go back to the
previous set of parameters and modify them a bit differently. Changing the parameters is discussed
in section 3.3.
The basic command to run ASP is:
stereo
--stereo-file
outputfolder/outputprefix
stereo.NACdefault
leftimage.cub
rightimage.cub
This uses your specified stereo parameter file (in this example called stereo.NACdefault) to
correlate your “leftimage” cube with your “rightimage” cube, creates an output folder named
outputfolder where it creates a large number of files, all of which have a file name starting with
outputprefix. It is useful to read about all of these files in the ASP guide book and learn to
understand at least the basics of them. The most straightforward of them is the outputprefixGoodPixelMap.tif file. Grey pixels are the good correlations, the red ones are the ones where the
correlation failed. This, and the number of holes you see given on the screen during the running of
stereo are the easiest ways to see if your DEM is going to be any good.
DEMs for dummies
Page 9 of 30
T. Öhman
6/21/2013
The most important file created at this stage is the outputprefix-PC.tif. Don’t bother trying to open
it. It’s a point cloud file, and normal image processors can’t handle it. But that is what you need in
creating the actual DEM. If the GoodPixelMap.tif looks good, you can skip the next section and
move straight into creating the DEMs instead, but if there are plenty of holes and mismatches, you
will need to change the correlation parameters in the stereo.default file.
3.3. Changing the correlation parameters
Although the default stereo parameter files that come with ASP and can also be found in the ASP
guidebook often work reasonably well for LRO NAC and CTX, there is no such file available for
HRSC, and in practice the parameters also for LRO NAC and CTX need to be changed to get the
best results. To do so, the stereo.default file has to be opened in a text editor. The easiest thing to do
is to use
gedit
The most important things to modify are the template window size, taken from the left image, and
the search range in the right image, from which ASP is trying to find the match. The idea should
become clearer from Figure 1, taken from the ASP guidebook (Broxton et al. 2012).
Figure 1. Correlation kernel (called Template Region in the figure) size is defined by parameters H_KERNEL
and V_KERNEL in the stereo file. Parameters H_CORR_MIN, H_CORR_MAX, V_CORR_MIN and
V_CORR_MAX define the correlation window (the search area). From Broxton et al. (2012).
DEMs for dummies
Page 10 of 30
T. Öhman
6/21/2013
When you open the stereo parameter file, you’ll find something like this in it about two thirds down
the file:
# Initialization step: correlation kernel size
H_KERNEL 7
V_KERNEL 7
# Initialization step: correlation window size
H_CORR_MIN -500
H_CORR_MAX 275
V_CORR_MIN -175
V_CORR_MAX 175
# Subpixel step: subpixel modes
#
# 0 - disable subpixel correlation (fastest)
# 1 - parabola fitting (draft mode - not as accurate)
# 2 - affine adaptive window, bayes EM weighting (slower, but much
more accurate)
SUBPIXEL_MODE 2
# Subpixel step: correlation kernel size
SUBPIXEL_H_KERNEL 50
SUBPIXEL_V_KERNEL 50
Correlation kernel (called template region in Figure 1) is the part of the left image for which ASP is
trying to find a counterpart in the right image. Its size is defined by parameters H_KERNEL and
V_KERNEL in the stereo file. Parameters H_CORR_MIN, H_CORR_MAX, V_CORR_MIN and
V_CORR_MAX define the correlation window, or the search area in the right image. In this
example the correlation kernel is unusually small, the more common values being around 29 or so,
but this is just an example where difficult areas failed with more standard settings. In contrast, the
correlation window size is here somewhat larger than in the default settings.
For refining the results, you may want to change the subpixel correlation size. Here it is larger than
usually, the default setting value being 25. These examples should give some idea what sort of
values may be useful. It is good to keep some track of what settings have worked in the past, so you
can easily use the same settings if you think the problem may be of similar type. If you don’t want
to make a huge number of different stereo parameter files, it is useful to know that one of the files
created by ASP in the outputfolder is called outputprefix-stereo.default, which is the parameter file
used to create the point cloud.
Full example stereo parameter files used in creating DEMs from LRO NAC, CTX, and HRSC
images are given in Appendices 3–5, respectively.
DEMs for dummies
Page 11 of 30
T. Öhman
6/21/2013
3.4. Creating the DEMs
3.4.1. The Moon
The point cloud file is not yet the actual DEM. That is created with point2dem command. There
are several different ways to use it for different purposes. Martian DEMs, if you want to compare
them with other datasets, are somewhat more complicated to produce and are discussed separately
in section 3.4.4. below. The Moon is an easier case, so to create a lunar DEM, the following works:
point2dem -r moon outputprefix-PC.tif -n
This creates a DEM (outputprefix-DEM.tif) assuming a spherical Moon with a radius of 1737.4 km,
which is what you would normally use. The switch -n is not necessary, but it does create a smaller
normalized image (outputprefix-DEM-normalized.tif) of the DEM for a quick check. This is handy,
because for example Photoshop does not easily understand the actual DEM.
For georeferencing in ArcMap (see section 4.4.) and for a number of other purposes, you do want to
have your image to match the DEM. This is done with point2dem as well:
point2dem -r moon --orthoimage outputprefix-L.tif outputprefix-PC-tif
This overlays the orthoprojected left image (outputprefix-L.tif) of your stereo pair to exactly match
the dimensions of the DEM. The resulting image (outputprefix-DRG.tif) will have all the holes that
the DEM does, so I only use that in ArcMap for georeferencing the DEM. For the actual image
analysis (e.g., mapping) after the georeferencing is done I use the original image cube, not the
*DRG.tif produced here.
If you have reached this point by working on the cropped test image, skip the next section and go
straight to the visualisation section 3.4.3.
3.4.2. Exporting the DEM to ArcGIS
To be able to use the final DEM and the orthoprojected image (DRG) in ArcGIS, you need to use
gdal for exporting them:
gdal_translate -ot float32 outputprefix-DEM.tif outputprefix-DEM_your_gdal_file.tif
gdal_translate -ot float32 outputprefix-DRG.tif outputprefix-DRG_ your_gdal_file.tif
These resulting files are now ready to be used in ArcGIS, discussed in chapter 4. However, a couple
of visualisation steps might be useful, so those are described next.
3.4.3. Visualising the DEM in ASP
Nothing described in this section is necessary for the analysis of DEMs, so this is for visualisation
purposes only. However, for seeing if a cropped test DEM is any good and, therefore, if it makes
sense to use the same parameters for the full stereo pair, these visualisation tools can give additional
easily understandable information about the DEM quality. The good pixel map may not always tell
the whole truth, and other debugging images provided by ASP and described in the guidebook are
not that simple to understand at first.
DEMs for dummies
Page 12 of 30
T. Öhman
6/21/2013
To see how the DEM really looks like without having to go through all the trouble with ArcGIS,
ASP has some handy visualisation tools. You can easily create hillshades, colour maps, coloured
hillshades, and perhaps most importantly, perspective views. Note that these are only for
visualisation purposes, the actual research on the DEM has to be done in ArcGIS. To create a
simple colour-coded DEM (blue=low, red=high) instead of the greyscale normalized DEM created
above, run this:
colormap outputprefix-DEM.tif -o outputprefix-DEM_color.tif
Making a hillshade is equally easy:
hillshade outputprefix-DEM.tif -o outputprefix-DEM_shade.tif -e 30
The -e switch defines the elevation of the light source, in this case 30º (0º being logically the
horizon). You can also change the azimuth of the light source, but that works in a less logical
manner. The default setting produces illumination from the left, and the azimuth increases from
there anti-clockwise. Therefore, using switches -e 5 -a 90 produces illumination from the bottom
of the image at a shallow angle of 5º , and -e 60 -a 225 more higher angle illumination from the
upper right, and so on.
You can also colourise your hillshade:
colormap outputprefix-DEM.tif
DEM_colorshade.tif
-s
outputprefix-DEM_shade.tif
-o
outputprefix-
The fanciest visualisation tool, however, uses the point cloud and the left image to create
perspective views that you can rotate freely, and even make your own flyovers (although that is
rather clumsy to do). Run the following:
point2mesh -l -s 5 outputprefix-PC.tif outputprefix-L.tif
Neither of the switches -l and -s is necessary, but both are recommendable. The first one gives you
the option of toggling additional illumination in the perspective view on or off. The second one, -s,
is a crucial one, as it defines the sampling rate. The default value is 10, which means it samples only
every 10th pixel. Depending on the size of your DEM, you probably want a higher sampling rate,
like every 5th pixel in this example. The limiting factor is the resulting file size. To see and play
with the perspective view, use the osgviewer:
osgviewer outputprefix.ive
In the osgviewer you can fly around using mouse, but the following keys are useful:
c: takes a screenshot (the same thing as hitting print screen)
h: allegedly displays help, but in practice they are tiny hieroglyphs on the screen
l: lighting on/off (if -l switch was used)
t: texture on/off (i.e. only the DEM-based surface model (mesh), or with the left image)
DEMs for dummies
Page 13 of 30
T. Öhman
6/21/2013
w: wire model on/off
z: starts recording the camera path
Z: stops recording the camera path and saves it to saved_animation.path
To playback the recorded camera path:
osgviewer outputprefix.ive -p saved_animation.path
While in the playback, hitting space bar sets you back to the beginning of the path, and hitting c
produces a short pause. Note that much of the information in the ASP guidebook about this part
seems not to work real life, at least using the settings at LPI.
ASP does not include the full OpenSceneGraph package. I have never used it, so I don’t know if it
has some smart way of exporting the perspective views, but the osgviewer included in ASP does
not. Therefore, the perspective views will be low resolution ones, but they can still be used in
papers as long as you don’t tell them they have a low resolution, and add some air in them in
Photoshop. Examples of these perspective views can be seen in the cover page and in Figures 2–4 in
the Appendices.
If the DEM looks good in whatever visualisation you prefer (to me the perspective views are by far
the most useful ones), you can export it to ArcGIS (chapter 3.4.2), or if you are still testing with a
cropped stereo pair, go back to cropping (chapter 3.1) and make the full-sized versions of the
images if you don’t have them already, and then go through the whole ASP process again and
repeat it as many times as necessary.
3.4.4. Martian DEMs and the areoid
To create Martian DEMs and to compare them with MOLA or DLR DEM (which are referenced to
MOLA) data, you need several additional steps. Run the following commands, which assume a
spherical Mars with a radius of 3396 km (that’s the MOLA spherical Mars, different from the IAU
Mars spheroid given by the option -r mars, and also different from the IAU Mars 2000 ellipsoid):
point2dem --semi-major-axis 3396000 --semi-minor-axis 3396000 outputprefix-PC.tif
-n
point2dem --semi-major-axis 3396000 --semi-minor-axis 3396000 --orthoimage
outputprefix-L.tif outputprefix-PC.tif
Then the complicated part begins. First you have to download the 16 pixels per degree (3.705
km/px) areoid file mega90n000eb.img and the accompanying label file mega90n000eb.lbl from:
http://pds-geosciences.wustl.edu/mgs/mgs-m-mola-5-megdr-l3-v1/mgsl_300x/meg016/
I would make a separate directory just for the various areoid files that will be created.
The areoid must of course first be ISISed to a cube using:
raw2isis
DEMs for dummies
Page 14 of 30
T. Öhman
6/21/2013
Open the label file in a text editor. In the label file, SAMPLES are called “LINE_SAMPLES”, and
there are 5760 of them. LINES are 2880. Although the label doesn’t directly say it, the BITTYPE is
16 bit signed data (SIGNEDWORD). BYTEORDER is found in the label, and it is Most Significant
Byte (MSB). Run it, and open the resulting cube in qview to make sure it looks alright (dark close to
the poles, brighter near the equator, Olympus Mons visible as a bright spot).
Create a maptemplate for the areoid:
maptemplate
In MAP type the name of this new map template you are about to create. In PROJECTION, select
Simple Cylindrical Projection (SIMPLECYLINDRICAL). Under Projection Parameters, type “180”
for the central longitude CLON. In TARGOPT, choose User Input (USER) and type “mars”. Keep
the other Target parameters at default values, but you can change both EQRADIUS and
POLRADIUS to 3396000 if you want to. However, these can be left to IAU Default settings as well,
because when no ground control points and bundle adjustment are used, in my test case on Alba
Mons this resulted in ~5 m elevation difference between the very final DEMs in ArcMap, while the
difference to the MOLA point elevation was ~110 m. So as long as you are not making comparisons
between different DEMs, whether you use the MOLA sphere (3396000 m) or the IAU defaults for
the radii, in the end it makes no practical difference one way or the other.
Under Ground Range Parameters, select User Entered Values (USER) for RNGOPT. The areoid is
global, so MINLAT is “-90”, MAXLAT “90”, MINLON “0”, and MAXLON “360”. Under
Resolution Parameters, choose RESOPT to be User Entered Scale (PPD), type “16” for
RESOLUTION, and run it.
Next the areoid and the mapping information in the map template have to be combined. This is done
by adding mapping labels to the areoid:
maplab
FROM is your areoid, and MAP is the newly created map template. Both SAMPLE and LINE are
0.5, which means the top left corner of the top left pixel. For COORDINATES, select Allows LAT,
LON as Input Values (LATLON). For LAT, type “90”, and for LON, type “0”. After creating the
mapping labels, it’s useful to check with tracking in qview that the coordinates really are correct in
the areoid. The bright dot of Olympus Mons has latitude of about 18ºN and longitude 226ºE.
The previous two steps, maptemplate and maplab, you only have to do once for the global areoid.
Once it’s done, you can keep on using it as a basis for all of your DEMs.
Now you need to crop the global areoid to approximately match the DEM. What I do, is I open the
left image of the stereo pair in qview and record the latitude and longitude of the top left corner of
the top left pixel and the bottom right corner of the bottom right pixel. Because of undesirable edge
effects in the next stage, it is best to add some margin. I usually use 0.05º margins. So do the math
for the coordinates recorded for the left image of the stereo pair, for example like this:
Top left longitude -0.05º
Top left latitude +0.05º
Bottom right longitude +0.05º
Bottom right latitude -0.05º
DEMs for dummies
Page 15 of 30
T. Öhman
6/21/2013
Then open the global areoid in qview, and use tracking to find the coordinates (including the
margins) of the DEM (i.e. the coordinates of the left image of the stereo pair recorded above), and
record (write down, really) the Sample and Line for the top left and bottom right corners. For top
left coordinates, round both Sample and Line down to the nearest integer, and for the bottom right
coordinates round them up (again just to play it safe with the margins, and in the end you can’t use
decimals in Sample and Line anyway).
Based on the lines and samples, calculate the size (subtract the top left samples and lines from the
bottom right ones) of the areoid to be cropped:
crop
Cropping is done same as before: FROM is the global areoid, TO would be smart to name it to be
specific for this DEM, SAMPLE and LINE are the top left sample and line recorded above,
NSAMPLES and NLINES are the size of the areoid calculated above, and SINC and LINC are 1.
Because of the enormous resolution difference between the 3705 m/px areoid and the ~6 m/px CTX
(and whatever HRSC resolution happens to be), the cropped areoid has to be scaled up to be about
the same fake resolution as the DEM. First calculate the scaling factor by dividing the areoid
resolution with the DEM resolution, for example 3705 / 6 = ~618. The scaling up is done with:
enlarge
FROM is naturally the cropped areoid, TO would benefit from a new name. Under Scaling
Parameters, choose INTERP to be Bi-linear interpolation (BILINEAR), as in my experience it gives
the best results in these types of situations. Cubic convolution would be another good option, but
nearest neighbor doesn’t work in this case. MODE should be Enlarge by a scaling factor (SCALE),
and both SSCALE and LSCALE should be, in this example, “618”. After running it, it’s good to
once again check it in qview. The left and right edge of the upscaled areoid will look coarser than
the rest of it, but that’s not a problem, as it fits into the margins. And there is also likely going to be
some “washboarding” in the areoid, but that’s just how it is. Record the latitude and longitude of the
top left and bottom right coordinates of the enlarged areoid.
This final areoid.cub (along with the gdaled *-DEM.tif and the accompanying gdaled *-DRG.tif)
can now be copied (just drag and drop from the SSH File Transfer Window, that’s the easiest way)
to the folder where your ArcGIS project is (or is going to be).
4. Processing and analysing the DEMs in ArcMap
4.1. Importing and re-projecting the data
Whether or not you already have a project in ArcMap, it doesn’t really matter. As always in
ArcMap, you should add the new datasets using the Add Data button. Start by adding the areoid
(assuming you are working with Martian data, which is the default assumption here – in the lunar
case everything is otherwise the same, just skip the areoid parts), DEM, and DRG. If and when Arc
wants to build pyramids, OK that. It may also give a Geographic Coordinate Systems Warning (if
you have an existing project, it will), but you can just ignore it for now, because it will be fixed
DEMs for dummies
Page 16 of 30
T. Öhman
6/21/2013
later. You should have three grey boxes now, one slightly larger (the areoid) and other two identical
ones inside it (not necessarily in the center of the areoid).
ArcGIS, of which ArcMap is the most commonly used part, is your worst enemy, so some routines
are useful, just to make sure things get always done the right way, and simply to make life a little
easier. Even if Arc didn’t complain now about the projection, it will eventually (or you just won’t
be able to georeference anything, which is more likely), which is why I always reproject every
single dataset that I import to my Arc projects. And I do it before I really do anything else.
To start with, try to locate a Search window. Type “project raster” in it. It should find Project
Raster (Data Management) tool. The chances are Arc will crash when you click it. It happens. It
just did to me right now. It’s not a bug, it’s just a fancy feature of Arc. So saving the project
immediately after importing the datasets would have been a smart thing to do… Once Project
Raster chooses to work, the steps for each file are the same: Input Raster is your input file (areoid,
DEM, or DRG). Output Raster Dataset only accepts file names up to 13 characters. You can give
them longer apparent names in the table of contents later. Arc is not always (ever?) too logical
about the default folder locations, so be sure you know where your output raster will be.
In Output Coordinate System, you should follow this path: Spatial Reference Properties  Select
 Geographic Coordinate Systems  Solar System  Mars  Mars 2000.prj. For lunar work,
Moon 2000.prj can be found under Earth. Then just Add and OK. Back in the Project Raster
window, you should now see the Output Coordinate System to be GCS_Mars_2000. The rest of the
settings can be defaults. Repeat this to all of the datasets.
While projecting, ArcMap may have already stretched the rasters so you can actually see them. If
not, right-clicking on the file in the Table of Contents gives you the option of fiddling with
Properties, although it is not that important yet if you are working with Martian data. If you’re
working with lunar data, jump right to the section 4.3. about Displaying the DEM.
4.2. Subtracting the areoid
Using the Search window, find the tool Minus (Spatial Analyst). ArcMap may crash, as it again just
did for me. When it eventually works (it may crash several times; mine just crashed for the third
time during the writing of this chapter 4), Input raster or constant value 1 is your DEM, Input
raster or constant value 2 is the areoid, and Output raster is the final DEM that has been the goal of
this whole lengthy exercise. Note that because no ground control points or bundle adjustment were
used in the creation of the DEM, the elevations will still be different from the MOLA point values,
which are the closest approximation of “the truth”. However, the elevations of the DEM are still
comparable to MOLA points, they are merely offset by some fairly random value. Usually the DEM
and MOLA point elevations are within a couple of hundred meters. Based on all the tests that I have
run, the research I’ve carried out and what I have read, DEMs created this way are a reliable source
of detailed topographic information as long as one does not try to compare the elevations between
different DEMs. For sanity checks, always use MOLA points (obviously LOLA in the lunar case),
photogeology, and common geologic sense.
4.3. Displaying the DEM
In the Table of Contents, the rasters can be moved up and down by simply grabbing them with the
mouse. You’ll want the DEM to be on top of the DRG. Right-clicking the name of the file in the
DEMs for dummies
Page 17 of 30
T. Öhman
6/21/2013
Table of Contents gives you the option of changing the Properties. In the Symbology tab, Color
Ramp is something you very likely want to change. The traditional blue-to-red ramp is available in
the lower part of the pull-down menu. If you use it, remember to check the Invert box at lower right.
The type of stretch varies between each individual case, but a good starting point is Standard
Deviations, with n being “2”. In the Display tab, you will likely want to change the Transparency
to somewhere between 40 and 60%. During this process, ArcMap will likely ask for your
permission to calculate statistics, which is what you want it to do. Now you should have your
partially transparent DEM overlaid exactly on top of your DRG, and basically it is ready for
analysis. However, if your project includes some background data, for example a WAC image or
mosaic, THEMIS mosaic, or MOLA DEM, it is likely that your DEM and DRG are slightly offset,
and need to be georeferenced.
4.4. Georeferencing
Normally you want to have some sort of a background layer that you presume to represent the
immovable reference framework. Some of them are closer to “the truth” than others, but for
example for Mars, THEMIS 100 m/px infrared daytime mosaic is a pretty good option (see
Appendix 1 for the URL). Assuming you have such a background, and your DEM, DRG, and/or
your original image file (cube, tif, jpeg2000…) are in the wrong place, you need to use the
Georeferencing toolbar. If it is not visible, choose it from Customize  Toolbars.
In the Georeferencing toolbar you should have all your rasters available in the Layer menu. If they
are not, you need to run Project raster on the missing ones again. ArcGIS 10 is particularly stupid
in this sense, hence my routine of re-projecting every raster I add immediately after I import it even
though it is supposedly already in the correct datum and projection.
I have so far met Martian and lunar ASP DEMs that have been only offset, and none have required
warping. Thus, no control points are required, and georeferencing is easy. Choose the raster you
want to move to the correct place from the Layer pull-down menu. The icon to the right of it usually
shows Rotate as a default, but the pull-down menu allows you to choose Shift, which is what you
want. This changes the cursor to a hand, which is very confusing, because it looks exactly the same
as the hand of the Pan tool that you would normally use to move around. With this Shift hand, move
the raster you want to the correct location. If everything is offset with respect to everything else
(which is quite likely), for example in the case of the Moon I would first move the NAC cube to
match the WAC background and georeference it (see below), then move the DRG (which, as a
reminder, is the left NAC cube projected to match exactly the DEM) to match the NAC, and finally
move the DEM to match the DRG. This last part can be done very precisely on sub-pixel level,
because the spatial extent of the DEM and the DRG are the same, so if you zoom in close enough,
you can exactly match the corners of the DEM and the DRG.
Once you have moved your raster-to-be-georeferenced to the right location, choose Rectify from the
Georeferencing pull-down menu. This opens up a Save as window. As usual, double-check the
Output Location, because it is likely going to be inherited from the previous project where
georeferencing was used. Give the raster a new Name, and save it.
Although the raster you moved appears to be in the place where you moved it to, it is not
permanently there, so you have to use the Add Data button to include the newly georeferenced
raster in your project. And just for the fun of it, you also have to stretch it again if you want it to
look the same as the original. Repeat this for all of the rasters you need to georeference. Depending
DEMs for dummies
Page 18 of 30
T. Öhman
6/21/2013
on the type of data, Rectify may not always work. In such a case you need to use Update
Georeferencing instead.
4.5. Analysing the DEM and exporting the data
Once all the rasters are in their correct locations, you can actually start analysing the DEM. There
are countless ways of doing that of course and I know virtually nothing about most of them, but
taking profiles is extremely useful and something that is almost always required. For that you need
the 3D Analyst toolbar. Again, if it is not visible, choose it from Customize  Toolbars. Choose the
Layer you want analyse, and draw your profile using the Interpolate Line button (the one with the
zigzag icon). First click starts the profile, subsequent clicks create points where you might want to
make a turn in the profile, and double-clicking ends the profile. Note that you cannot use the same
profile for more than one layer. So if you need to take the same profile on several different rasters,
you need to draw the profile separately for every raster. For precise work, zoom in a lot so you are
at least in the same pixel for each of the profiles.
Create Profile Graph button on the right, surprisingly enough, usually creates the profile graph.
When it does, right-click on the graph. Choose Export from the menu that appears, select Data tab,
choose the format you like, and then just Save. In the resulting Excel-file (or whatever you want to
use), the Y column displays the elevation from your DEM in meters. The X column is normally a
bit more problematic, because it tends to be given in degrees. There is probably some way to solve
this in ArcMap, but I don’t know what it is. Therefore, I use the Measure tool (button with a ruler
icon) to measure the length of the profile, and then in Excel I use the degrees to calculate the Xvalues in kilometers for the points between the starting and ending points of the profile.
It is worth noting that right-clicking the profile gives the option of seeing its Properties. In the
Length tab, coordinates of the starting and ending points are correct and useful information, whereas
the length of the profile there is practically always wrong (There are sort of sensible reasons why
this is so. For the same reasons, inserting a scale bar when creating a map is not advisable, because
it will be wrong.). So to get the true length of the profile, you have to use the Measure button.
And that is pretty much all there is to it from a processing point of view. The rest is just doing the
analysis and trying to make some sense of it all.
Acknowledgements
This project was supported by NASA Mars Data Analysis Program (MDAP) grant NNX09AI42G
(PI Patrick J. McGovern), and earlier kinks in the learning curve were funded by NASA Lunar
Science Institute contract NNA09DB33A (PI David A. Kring), and by NASA under the LPI
Cooperative Agreement NNX08AC28A. A big thank you to Brian Fessler for all his help with ISIS
and ArcGIS, and setting up ASP.
DEMs for dummies
Page 19 of 30
T. Öhman
6/21/2013
Appendix 1. Some additional links and references
Planetary datasets, and particularly any useful information about using them, are notoriously
difficult to find. Here just some of the sites that I have found useful.
 The reference to the ASP guidebook is:
Broxton M. J., Beyer R. A., Moratto Z., Lundy M., Husman K., 2012. The Ames Stereo Pipeline:
NASA’s Open Source Automated Stereogrammetry Software. A part of the NASA NeoGeography
Toolkit. Version 2.0.0. Intelligent Robotics Group, NASA Ames Research Center, 88 p.
 A couple of LPSC abstracts that the ASP developers want to be referenced are:
Broxton M. J., Edwards L. J., 2008. The Ames Stereo Pipeline: automated 3D surface
reconstruction from orbital imagery, Abstract #2419, 39th Lunar and Planetary Science Conference.
http://www.lpi.usra.edu/meetings/lpsc2008/pdf/2419.pdf
Moratto Z. M., Broxton M. J., Beyer R. A., Lundy M., Husmann K., 2010. Ames Stereo Pipeline,
NASA’s open source automated stereogrammetry software. Abstract #2364, 41st Lunar and
Planetary Science Conference.
http://www.lpi.usra.edu/meetings/lpsc2010/pdf/2364.pdf
 One paper where I know a similar approach as presented here (no ground control points, no
bundle adjustment) was used, with the author list even including one of the ASP developers, is:
Lefort A., Burr D. M., Beyer R. A., Howard A. D., 2012. Inverted fluvial features in the AeolisZephyria Plana, western Medusae Fossae Formation, Mars: Evidence for post formation
modification. Journal of Geophysical Research, 117, E03007, doi: 10.1029/2011JE004008.
 ASP home page, with links to the ASP guidebook and other useful things, including the
invaluable email list:
http://ti.arc.nasa.gov/tech/asr/intelligent-robotics/ngt/stereo/
 ASP-developer Zach Moratto’s blog with many useful tips:
http://lunokhod.org/
 ISIS guidebook:
http://isis.astrogeology.usgs.gov/Application/index.html
 ISIS discussion forum:
https://isis.astrogeology.usgs.gov/IsisSupport/
 USGS guide on Working with Mars Reconnaissance Orbiter CTX Data:
DEMs for dummies
Page 20 of 30
T. Öhman
6/21/2013
http://isis.astrogeology.usgs.gov/IsisWorkshop/index.php/Working_with_Mars_Reconnaissance_Or
biter_CTX_Data
 GDAL conversion tips from Trent Hare:
https://isis.astrogeology.usgs.gov/IsisSupport/index.php/topic,2172.0.html
 What is SPICE:
http://naif.jpl.nasa.gov/naif/spiceconcept.html
 Global THEMIS IR daytime 512 ppd (100 m/px) mosaic:
ftp://pdsimage2.wr.usgs.gov/pub/pigpen/mars/themis/themis_ir_global_mosaic/Global_IRday_512p
pd/
 LOLA points, to compare with lunar stereo DEMs:
http://ode.rsl.wustl.edu/moon/indextools.aspx
 LOLA DEM in simple cylindrical projection and jpeg2000 format:
http://imbrium.mit.edu/DATA/LOLA_GDR/CYLINDRICAL/JP2/
 LRO WAC 128 px/degree stereo DEM:
http://wms.lroc.asu.edu/lroc/global_product/color_shaded_relief_grid
 MOLA points, to compare with Martian stereo DEMs:
http://ode.rsl.wustl.edu/mars/indextools.aspx?displaypage=molapedr
 MOLA DEMs and areoid, as well as radius data:
http://pds-geosciences.wustl.edu/missions/mgs/megdr.html
 CTX home page:
http://www.msss.com/all_projects/mro-ctx.php
 LROC home page:
http://www.lroc.asu.edu/
DEMs for dummies
Page 21 of 30
T. Öhman
6/21/2013
Appendix 2. Some very basic UNIX/LINUX commands
cd directoryname
Goes to a directory.
cd ..
Goes up one level in directories.
cp oldfilename newfilename
Copies a file.
exit
A sophisticated way of quitting. Note that you will lose your
terminal history (available with the up-arrow) when you exit.
gedit
Starts a text editor.
ls
Lists the files in the current directory.
mkdir directoryname
Makes a new directory.
more filename
Shows the contents of the file on the screen, stopping when
the screen is full. Hitting Ctrl-z stops it.
rm filename
Removes (deletes) the file.
rmdir directoryname
Removes (deletes) an empty directory.
DEMs for dummies
Page 22 of 30
T. Öhman
6/21/2013
Appendix 3. An example of a stereo parameter file for processing LRO NAC stereo pairs
These parameters were used to produce nice DEMs of the central part of crater Eimmart A from
LRO NAC images M152451994LE.IMG and M152445210LE.IMG.
#############################
##
PREPROCESSING
##
#############################
# Pre-alignment options
DO_INTERESTPOINT_ALIGNMENT 0
INTERESTPOINT_ALIGNMENT_SUBSAMPLING 0
DO_EPIPOLAR_ALIGNMENT 0
# Normalization options
FORCE_USE_ENTIRE_RANGE 1
DO_INDIVIDUAL_NORMALIZATION 0
#
#
#
#
#
#
Preprocessing filter mode:
0
1
2
3
-
None
Gaussian Blur
Laplacian of Gaussian Blur (recommended)
Sign of the LoG Filter
PREPROCESSING_FILTER_MODE 2
# Gaussian convolution kernel size (1-sigma) for pre-processing
# modes 1, 2 and 3.
SLOG_KERNEL_WIDTH 1.5
#########################################################################
###########################
CORRELATION
###########################
#########################################################################
#
#
#
#
#
Select a cost function to use for initialization:
0 - absolute difference (fast -- use in conjunction with SLoG)
1 - squared difference
2 - normalized cross correlation (recommended)
COST_MODE 2
#
#
#
#
#
#
#
#
Turn this up to improve the results of the discrete correlation
step. This will reduce the number of missing pixels, but reduce the
overall accuracy of the disparity estimates.
WARNING: Cost blurring must be used in conjunction with affine
adaptive window subpixel below, which are capable of achieving
highly accurate results even when seeded by slightly inaccurate
results from the initialization step.
COST_BLUR 12
DEMs for dummies
Page 23 of 30
T. Öhman
6/21/2013
# Initialization step: correlation kernel size
H_KERNEL 7
V_KERNEL 7
# Initialization step: correlation window size
H_CORR_MIN -500
H_CORR_MAX 275
V_CORR_MIN -175
V_CORR_MAX 175
# Subpixel step: subpixel modes
#
# 0 - disable subpixel correlation (fastest)
# 1 - parabola fitting (draft mode - not as accurate)
# 2 - affine adaptive window, bayes EM weighting (slower, but much more
accurate)
SUBPIXEL_MODE 2
# Subpixel step: correlation kernel size
SUBPIXEL_H_KERNEL 25
SUBPIXEL_V_KERNEL 25
#########################################################################
############################
FILTERING
############################
#########################################################################
# Fill in holes up to 100,000 pixels in size with an inpainting method
FILL_HOLES 1
# Automatic "erode" low confidence pixels
RM_H_HALF_KERN 5
RM_V_HALF_KERN 5
RM_MIN_MATCHES 60
RM_THRESHOLD 3
RM_CLEANUP_PASSES 1
# Note: a 5*5 halfkernel = 11*11 kernel -> 121 pixels
# Units = percent
#########################################################################
#########################
POST_PROCESSING
#########################
#########################################################################
# Size max of the universe in meters and altitude off the ground.
# Setting both values to zero turns this post-processing step off.
NEAR_UNIVERSE_RADIUS 0.0
FAR_UNIVERSE_RADIUS 0.0
DEMs for dummies
Page 24 of 30
T. Öhman
6/21/2013
Figure 2. A perspective view of Eimmart A created with ASP using NAC images and stereo parameters
described above. Diameter of the crater is ~7 km.
DEMs for dummies
Page 25 of 30
T. Öhman
6/21/2013
Appendix 4. An example of a stereo parameter file for processing CTX stereo pairs
These parameters were used to produce nice DEMs of the graben on the eastern flank of Alba Mons
from
CTX
images
G20_025970_2217_XN_41N102W.IMG
and
G20_025904_2209_XN_40N102W.IMG. See the cover page for an example and comparisons with
other datasets.
#############################
##
PREPROCESSING
##
#############################
# Pre-alignment options
DO_INTERESTPOINT_ALIGNMENT 1
INTERESTPOINT_ALIGNMENT_SUBSAMPLING 0
DO_EPIPOLAR_ALIGNMENT 0
# Normalization options
FORCE_USE_ENTIRE_RANGE 1
DO_INDIVIDUAL_NORMALIZATION 0
#
#
#
#
#
#
Preprocessing filter mode:
0
1
2
3
-
None
Gaussian Blur
Laplacian of Gaussian Blur (recommended)
Sign of the LoG Filter
PREPROCESSING_FILTER_MODE 3
# Gaussian convolution kernel size (1-sigma) for pre-processing
# modes 1, 2 and 3.
SLOG_KERNEL_WIDTH 1.5
#########################################################################
###########################
CORRELATION
###########################
#########################################################################
#
#
#
#
#
Select a cost function to use for initialization:
0 - absolute difference (fast)
1 - squared difference (faster .. but usually bad)
2 - normalized cross correlation (recommended)
COST_MODE 0
#
#
#
#
#
#
#
#
Turn this up to improve the results of the discrete correlation
step. This will reduce the number of missing pixels, but reduce the
overall accuracy of the disparity estimates.
WARNING: Cost blurring must be used in conjunction with affine
adaptive window subpixel below, which are capable of achieving
highly accurate results even when seeded by slightly inaccurate
results from the initialization step.
DEMs for dummies
Page 26 of 30
T. Öhman
6/21/2013
COST_BLUR 0
# Initialization step: correlation kernel size
H_KERNEL 25
V_KERNEL 25
# Initializaion step: correlation window size
#H_CORR_MIN -100
#H_CORR_MAX 100
#V_CORR_MIN -100
#V_CORR_MAX 100
# Subpixel step: subpixel modes
#
# 0 - disable subpixel correlation (fastest)
# 1 - parabola fitting (draft mode - not as accurate)
# 2 - affine adaptive window, bayes EM weighting (slower, but much more
accurate)
SUBPIXEL_MODE 2
# Subpixel step: correlation kernel size
SUBPIXEL_H_KERNEL 25
SUBPIXEL_V_KERNEL 25
#########################################################################
############################
FILTERING
############################
#########################################################################
# Fill in holes up to 100,000 pixels in size with an inpainting method
FILL_HOLES 1
# Automatic "erode" low confidence pixels
RM_H_HALF_KERN 5
RM_V_HALF_KERN 5
RM_MIN_MATCHES 60
RM_THRESHOLD 3
RM_CLEANUP_PASSES 1
# Note: a 5*5 halfkernel = 11*11 kernel -> 121 pixels
# Units = percent
#########################################################################
#########################
POST_PROCESSING
#########################
#########################################################################
# Size max of the universe in meters and altitude off the ground.
# Setting both values to zero turns this post-processing step off.
NEAR_UNIVERSE_RADIUS 0.0
FAR_UNIVERSE_RADIUS 0.0
DEMs for dummies
Page 27 of 30
T. Öhman
6/21/2013
Figure 3. A perspective view of a central pit crater and graben in eastern Alba Mons created with ASP
using CTX images and stereo parameters described above. Diameter of the crater is ~19 km.
Another example can be seen on the cover page.
DEMs for dummies
Page 28 of 30
T. Öhman
6/21/2013
Appendix 5. An example of a stereo parameter file for processing HRSC stereo pairs
These parameters were used to produce reasonably nice DEMs of the graben on the eastern flank of
Alba Mons using HRSC images H0068_0009_S12.IMG and H0068_0009_S22.IMG.
DO_INTERESTPOINT_ALIGNMENT 0
INTERESTPOINT_ALIGNMENT_SUBSAMPLING 0
DO_EPIPOLAR_ALIGNMENT 0
FORCE_USE_ENTIRE_RANGE 1
DO_INDIVIDUAL_NORMALIZATION 0
PREPROCESSING_FILTER_MODE 2
SLOG_KERNEL_WIDTH 1.5
COST_MODE 0
COST_BLUR 0
H_KERNEL 22
V_KERNEL 22
# These numbers need to be customized for every HRSC stereo session
H_CORR_MIN -30
H_CORR_MAX 30
V_CORR_MIN -130
V_CORR_MAX 20
SUBPIXEL_MODE 2
SUBPIXEL_H_KERNEL 35
SUBPIXEL_V_KERNEL 35
FILL_HOLES 1
RM_H_HALF_KERN 5
RM_V_HALF_KERN 5
RM_MIN_MATCHES 60
RM_THRESHOLD 3
RM_CLEANUP_PASSES 1
# Note: a 5*5 halfkernel = 11*11 kernel -> 121 pixels
# Units = percent
NEAR_UNIVERSE_RADIUS 0.0
FAR_UNIVERSE_RADIUS 0.0
DEMs for dummies
Page 29 of 30
T. Öhman
6/21/2013
Figure 4. A perspective view of graben and central pit craters in eastern Alba Mons created with ASP using
HRSC images and stereo parameters described above. The width of the view is ~155 km, and the crater in
the center left is the same as in Figure 3.
DEMs for dummies
Page 30 of 30